World Psychiatry / Volume 22, Issue 1 / pp.

105-115

Research Reports Free Access

Cognitive behavior therapy vs. control conditions, other

psychotherapies, pharmacotherapies and combined
treatment for depression: a comprehensive meta-analysis
including 409 trials with 52,702 patients

Pim Cuijpers, Clara Miguel, Mathias Harrer, Constantin Yves Plessen,

Marketa Ciharova, David Ebert, Eirini Karyotaki

First published: 14 January 2023

https://doi.org/10.1002/wps.21069
Citations: 23

Abstract
Cognitive behavior therapy (CBT) is by far the most examined type of
psychological treatment for depression and is recommended in most
treatment guidelines. However, no recent meta-analysis has
integrated the results of randomized trials examining its e"ects, and
its e#cacy in comparison with other psychotherapies,
pharmacotherapies and combined treatment for depression remains
uncertain. We searched PubMed, PsycINFO, Embase and the
Cochrane Library to identify studies on CBT, and separated included
trials into several subsets to conduct random-e"ects meta-analyses.
We included 409 trials (518 comparisons) with 52,702 patients, thus
conducting the largest meta-analysis ever of a speci$c type of
psychotherapy for a mental disorder. The quality of the trials was
found to have increased signi$cantly over time (with increasing
numbers of trials with low risk of bias, less waitlist control groups,
and larger sample sizes). CBT had moderate to large e"ects

https://onlinelibrary.wiley.com/doi/full/10.1002/wps.21069 07/06/2025, 13 22
Page 1 of 32
:
 compared to control conditions such as care as usual and waitlist
(g=0.79; 95% CI: 0.70-0.89), which remained similar in sensitivity
analyses and were still signi$cant at 6-12 month follow-up. There was
no reduction of the e"ect size of CBT according to the publication
year (<2001 vs. 2001-2010 vs. >2011). CBT was signi$cantly more
e"ective than other psychotherapies, but the di"erence was small
(g=0.06; 95% CI: 0-0.12) and became non-signi$cant in most
sensitivity analyses. The e"ects of CBT did not di"er signi$cantly from
those of pharmacotherapies at the short term, but were signi$cantly
larger at 6-12 month follow-up (g=0.34; 95% CI: 0.09-0.58), although
the number of trials was small, and the di"erence was not signi$cant
in all sensitivity analyses. Combined treatment was more e"ective
than pharmacotherapies alone at the short (g=0.51; 95% CI: 0.19-
0.84) and long term (g=0.32; 95% CI: 0.09-0.55), but it was not more
e"ective than CBT alone at either time point. CBT was also e"ective
as unguided self-help intervention (g=0.45; 95% CI: 0.31-0.60), in
institutional settings (g=0.65; 95% CI: 0.21-1.08), and in children and
adolescents (g=0.41; 95% CI: 0.25-0.57). We can conclude that the
e#cacy of CBT in depression is documented across di"erent formats,
ages, target groups, and settings. However, the superiority of CBT
over other psychotherapies for depression does not emerge clearly
from this meta-analysis. CBT appears to be as e"ective as
pharmacotherapies at the short term, but more e"ective at the
longer term.

Depression is a highly prevalent mental disorder, with about 280 million

people worldwide su"ering from it1. The disorder results in considerable
loss of quality of life in patients and their families2, and is associated
with increased physical morbidity and premature mortality3, a
considerable disease burden at the population level1, and enormous
economic costs4. Several evidence-based interventions are available for
the treatment of depression, including pharmacotherapies5 and
6

https://onlinelibrary.wiley.com/doi/full/10.1002/wps.21069 07/06/2025, 13 22
Page 2 of 32
:
 psychotherapies6.

Cognitive behavior therapy (CBT) is by far the most examined type of

psychological treatment for depression and is recommended in most
treatment guidelines. Several hundreds of randomized controlled trials
have tested the e"ects of CBT6, 7. Previous meta-analyses have found
that CBT is signi$cantly more e"ective in the treatment of depression
than various control conditions6-8, whereas its e"ectiveness in
comparison with other psychotherapies, pharmacotherapies and
combined treatment at the short and longer term, as well as its impact
on speci$c populations of patients and in di"erent formats, remain
uncertain9.

The last comprehensive meta-analysis of CBT for depression was

published in 20138, while the number of trials has increased
exponentially over the years, and many new trials have been published
since then. Furthermore, that meta-analysis did not include trials in
children/adolescents and inpatients, as well as comparisons with
pharmacotherapies and combined treatments, with other
psychotherapies, and with unguided digital interventions. More recent
meta-analyses have focused on psychological interventions in general,
including CBT6, 7, but they have not examined speci$c characteristics of
the participants, the treatment and the study as predictors of outcome.

We decided, therefore, to conduct a new, comprehensive meta-analysis

of randomized trials examining the short- and long-term e"ects of CBT
in depression across all treatment formats (i.e., individual, group,
unguided and guided self-help), all ages (including children and
adolescents), delivered in any setting (including outpatients and
inpatients), and compared against control conditions (e.g., waitlist, care
as usual) as well as other active treatments (i.e., other psychotherapies,
antidepressant medications, and combined treatment).

https://onlinelibrary.wiley.com/doi/full/10.1002/wps.21069 07/06/2025, 13 22
Page 3 of 32
:
 METHODS
Identi$cation and selection of trials
This study is part of a larger meta-analytic project on psychological
treatments for depression10. The protocol for the current meta-analysis
has been published in the Open Science Framework
(http://osf.io/a6p3w).

The trials included in this study were identi$ed through a database

which is continuously updated, currently including studies from 1966 to
January 1, 2022. For this database, we searched PubMed, PsycINFO,
Embase and the Cochrane Library, by combining index and free terms
indicative of depression and psychotherapies, with $lters for
randomized controlled trials. The full search strings can be found in the
supplementary information. Furthermore, we checked references of
earlier meta-analyses on psychological treatments for depression.

Two independent researchers screened all records, and all papers that
could meet inclusion criteria according to one of them were retrieved as
full text. The two independent researchers also decided to include or
exclude a study in the database, and disagreements were resolved
through discussion.

For the current study, we selected randomized controlled trials in which

CBT for people with depression was compared with control conditions
(care as usual, waitlist, others), other psychotherapies,
pharmacotherapies, or combined treatment.

A broad de$nition of CBT was used: a treatment in which the therapist

focuses on the impact of present dysfunctional thoughts on a patient's
current behavior and future functioning, and which is aimed at
evaluating, challenging and modifying a patient's dysfunctional beliefs

https://onlinelibrary.wiley.com/doi/full/10.1002/wps.21069 07/06/2025, 13 22
Page 4 of 32
:
 (cognitive restructuring). Cognitive restructuring could be combined with
other mood management skills, such as behavioral activation, problem-
solving, social skills training, or mindfulness. This de$nition was derived
from an extensive study in which di"erent types of psychotherapies
were examined by multiple researchers, resulting in a consensus on the
de$nition of each therapy11.

Depression could be de$ned as meeting the criteria for a depressive

disorder according to a diagnostic interview or as a score above the cut-
o" on a self-report depression measure. We included trials in which CBT
was administered in any format (individual, group, telephone, guided or
unguided self-help). We also included trials of outpatients as well as
inpatients, and in any age group.

We separated the included studies into several subsets, so that the

comparisons from these studies could be pooled in a meta-analysis. In
the largest subset, CBT was compared with control conditions. In this
subset, we included CBT that was applied individually, in groups, as
guided self-help, or in a mixed format, because previous research has
shown that these formats have comparable e"ects12. Studies of
unguided self-help CBT were included in a separate subset. We also
created a separate subset for CBT in inpatients, because these patients
di"er from outpatients, and the control conditions vary considerably
from outpatient settings13. A separate subset was also built for studies
comparing CBT with pharmacotherapies, CBT with combined treatment,
and pharmacotherapies with combined treatment. We created a
separate subset for depression in children and adolescents, because
therapies usually are less e"ective in this group.

Quality assessment and data extraction

We assessed the validity of included studies using four criteria of the
Risk of Bias (RoB) assessment tool, version 1, developed by the Cochrane
14, 15

https://onlinelibrary.wiley.com/doi/full/10.1002/wps.21069 07/06/2025, 13 22
Page 5 of 32
:
 Collaboration14, 15. The RoB tool assesses possible sources of bias in
randomized trials, including the adequate generation of allocation
sequence; the concealment of allocation to conditions; the prevention of
knowledge of the allocated intervention (masking of assessors); and
dealing with incomplete outcome data (this was assessed as positive
when intention-to-treat analyses were conducted, meaning that all
randomized patients were included in the analyses). Two independent
researchers evaluated the validity of the included studies, and
disagreements were solved through discussion.

We also coded participant characteristics (diagnostic method,

recruitment method, target group, mean age, proportion of women,
inpatient or outpatient); characteristics of CBT (treatment format,
number of sessions), as well as general characteristics of the studies
(type of comparison group, publication year, country where the study
was conducted). In the studies in which CBT was compared with other
therapies, we also categorized the other therapies according to the
de$nitions provided elsewhere7. In studies with pharmacotherapies, we
also categorized the type of antidepressant: selective serotonin reuptake
inhibitor (SSRI), tricyclic antidepressant (TCA), other.

Outcome measures
For each comparison between a psychological treatment and a control
condition, the e"ect size indicating the di"erence between the two
groups at post-test was calculated (Hedges’ g)16. E"ect sizes were
calculated by subtracting (at post-test) the average score of the
psychotherapy group from the average score of the control group and
dividing the result by the pooled standard deviation. Because some
studies were expected to have relatively small sample sizes, we
corrected the e"ect size for small sample bias.

When the means and standard deviations were not reported in a study,

https://onlinelibrary.wiley.com/doi/full/10.1002/wps.21069 07/06/2025, 13 22
Page 6 of 32
:
 we used change scores. If these were not reported, we converted binary
outcomes to Hedges’ g. If these were also not reported, we used other
statistics (e.g., p value, t value) to calculate the e"ect size.

Meta-analyses
To make a historical overview of trials on CBT over time, we conducted
bivariable linear regression analyses examining if the characteristics of
the trials have changed over time. We limited these analyses to the
subset comparing CBT with control conditions, because this was the
largest and most homogeneous subset.

The meta-analyses were conducted using the metapsyTools package in R

(version 4.1.1) and Rstudio (version 1.1.463 for Mac)17. The metapsyTools
package was speci$cally developed for the meta-analytic project of
which this study is part. This package imports the functionality of the
meta18, metafor19, and dmetar20 packages.

We calculated the pooled e"ect sizes in several di"erent ways, as

implemented in the metapsyTools package, so that we could explore if
di"erent pooling methods resulted in di"erent outcomes. In our main
model, all e"ect size data available for a comparison in a speci$c study
were aggregated within that comparison $rst. These aggregated e"ects
were then pooled across studies and comparisons. An intra-study
correlation coe#cient of ρ=0.5 was assumed to aggregate e"ects within
comparisons.

We conducted several other analyses to examine whether these main

outcomes were robust. First, we estimated the pooled e"ect using a
three-level correlated and hierarchical e"ects (CHE) model21. We
assumed an intra-study correlation of ρ=0.5 for this model. Second, we
pooled e"ects while excluding outliers, using the “non-overlapping
con$dence intervals” approach, in which a study is de$ned as an outlier

https://onlinelibrary.wiley.com/doi/full/10.1002/wps.21069 07/06/2025, 13 22
Page 7 of 32
:
 when the 95% con$dence interval (CI) of the e"ect size does not overlap
with the 95% CI of the pooled e"ect size20. Third, we pooled e"ects while
excluding in%uential cases, de$ned by the diagnostics proposed by
Viechtbauer and Cheung22. Fourth, we calculated the e"ect when the
smallest or largest e"ect in each study was considered. Fifth, we
estimated the pooled e"ect using only studies with a low risk of bias. We
also used three di"erent methods to assess and adjust for potential
publication bias20, 23: Duval and Tweedie's trim and $ll procedure24,
Rücker's “limit meta-analysis method”25, and the selection model26, 27.

A random-e"ects model was assumed for all analyses. Between-study

heterogeneity variance (components) was estimated using restricted
maximum likelihood. For models not $tted using robust variance
estimation, we applied the Knapp-Hartung method to obtain robust CIs
and signi$cance tests of the overall e"ect28.

As a test of homogeneity of e"ect sizes, we calculated the I2-statistic and

its 95% CI, which is an indicator of heterogeneity in percentages. A value
of 0% indicates no observed heterogeneity, and larger values indicate
increasing heterogeneity, with 25% as low, 50% as moderate, and 75% as
high heterogeneity29. For the three-level model, we calculated a
multilevel extension of I2, which describes the amount of total variability
attributable to heterogeneity within studies (level 2) and heterogeneity
between studies (level 3)20, 30. Because I2 cannot be interpreted as an
absolute measure of the between-study heterogeneity, we also added
the prediction interval (PI) to the main analyses, which indicates the
range in which the true e"ect size of 95% of all populations will fall31, 32.

We also estimated the number-needed-to-treat (NNT) for depression

using the formulae provided by Furukawa33 (assuming the control
group's event rate at a conservative 17%)34.

For the main comparison (CBT versus control conditions), we also

https://onlinelibrary.wiley.com/doi/full/10.1002/wps.21069 07/06/2025, 13 22
Page 8 of 32
:
 extracted the rate of response (i.e., a 50% reduction of depressive
symptoms compared to baseline). If the response rate was not reported,
we estimated it using a method based on the baseline means, the post-
test means, the post-test standard deviations and the number of
subjects35. For studies using the Hamilton Rating Scale for Depression
(HAM-D), we also calculated the rate of remission, de$ned as a score of
≤7 on the 17-item version of that scale36. We also calculated the relative
risk (RR) for response and remission of CBT compared with the control
groups, as well as the NNT (as 1 divided by the risk di"erence).

In each subset, we conducted a series of subgroup analyses, examining

the e"ects of the interventions according to major characteristics of the
participants, interventions and studies. We avoided subgroups of less
than $ve studies, merging them with other subgroups. Because the
subset comparing CBT with control conditions was very large, we also
conducted a multivariable meta-regression analysis in which all
characteristics were included.

RESULTS
Selection and inclusion of studies
After examining a total of 30,889 records (21,563 after removal of
duplicates), we retrieved 3,584 full-text papers for further consideration.
A total of 409 trials met the inclusion criteria for this meta-analysis (see
Figure 1). Selected characteristics of included studies and comparisons
are presented in the supplementary information.

https://onlinelibrary.wiley.com/doi/full/10.1002/wps.21069 07/06/2025, 13 22
Page 9 of 32
:
 Figure 1 Open in !gure viewer PowerPoint

PRISMA %ow chart, CBT – cognitive behavior therapy

Characteristics of included studies

The 409 studies (518 comparisons between CBT and a control condition)
included 52,702 patients (27,000 in CBT and 25,702 in control groups).
Aggregated characteristics of the studies and comparisons are provided
in Table 1.

Table 1. Aggregated characteristics of included studies and comparisons

https://onlinelibrary.wiley.com/doi/full/10.1002/wps.21069 07/06/2025, 13 22
Page 10 of 32
:
 Included studies (n=409)

Recruitment, n (%) Community 181

(44.3)

Clinical 106
(25.9)

Other 122
(29.6)

Target group, n (%) Children 12 (2.9)

Adolescents 25 (6.1)

Adults 160
(39.1)

Elderly 26 (6.4)

General medical 70 (17.1)

Perinatal 41 (10.0)

Other 75 (18.3)

Age, years (mean±SD) 40.1±15.0

Gender (% female) 69.0

Diagnosis, n (%) Meeting criteria for depressive disorder 226

Most studies recruited participants through the community (n=181,

44.3%) or clinical referrals (n=106, 25.9%). In most studies, the target
group was represented by adults in general (n=160, 39.1%); 70 studies
aimed at patients with general medical disorders (17.1%), 41 studies at
perinatal depression (10.0%), and 27 studies at children or adolescents
(9.0%).

https://onlinelibrary.wiley.com/doi/full/10.1002/wps.21069 07/06/2025, 13 22
Page 11 of 32
:
 In the majority of studies (n=226, 55.3%), depression was de$ned as
meeting the criteria for a depressive disorder according to a diagnostic
interview, while in 162 studies (39.3%) it was de$ned as a score above
the cut-o" on a self-report depression measure. The mean age of
participants in the studies was 40.1±14.98 years; the average proportion
of women was 69%. Most studies were conducted in the US (n=141,
34.5%) or in the UK or other European countries (n=141, 34.5%). Most
studies (n=249, 60.8%) were published since 2011.

Among the 518 comparisons, the majority tested an individual CBT

format (n=206, 39.8%), while 141 examined a group format (27.2%), 84 a
guided self-help format (16.2%), and 39 an unguided self-help format
(7.5%). In 211 comparisons (40.7%), CBT was administered in more than
12 sessions.

Of the 409 studies, 224 (54.8%) reported an adequate generation of

allocation sequence, 201 (49.1%) an adequate concealment of allocation
to conditions, and 101 (24.7%) an adequate prevention of knowledge of
the allocated intervention (masking of assessors); 262 (64.1%) conducted
intention-to-treat analyses. Risk of bias was low across all four domains
in 131 studies (32.0%), for two or three domains in 173 studies (42.2%),
and for no or one domain in 105 studies (25.7%).

Historical overview
The historical overview was limited to the subset comparing CBT with
control conditions (241 studies with 271 comparisons, including 12,907
patients in CBT arms and 12,199 in control conditions). The cumulative
number of studies over time is shown in Figure 2.

https://onlinelibrary.wiley.com/doi/full/10.1002/wps.21069 07/06/2025, 13 22
Page 12 of 32
:
 Figure 2 Open in !gure viewer PowerPoint

Randomized trials comparing cognitive behavior therapy (CBT) with control

conditions: cumulation over time

The bivariable linear regression analyses found that the number of trials
examining depressed patients with general medical disorders and
women with perinatal depression increased signi$cantly over time
(p=0.007 and p=0.012, respectively). The use of waitlist as the control
condition decreased signi$cantly over time (p=0.001), while the number
of studies with low risk of bias increased signi$cantly (p<0.001), as well
as the number of trials in non-Western countries (p=0.005). The number
of participants in each comparison also increased signi$cantly (p<0.001),
while the number of sessions of CBT decreased signi$cantly over time
(p=0.03). All the other characteristics of CBT trials did not change over
time (see also supplementary information).

CBT versus control conditions

The main e"ect size indicating the overall di"erence between CBT and

https://onlinelibrary.wiley.com/doi/full/10.1002/wps.21069 07/06/2025, 13 22
Page 13 of 32
:
 control conditions after treatment was g=0.79 (95% CI: 0.70-0.89),
corresponding to an NNT of 3.8 (see Table 2). Heterogeneity was very
high (I2=85; 95% CI: 83-86), and the prediction interval ranged from –0.45
to 2.04.

Table 2. Cognitive behavior therapy (CBT) vs. control conditions: main

analyses

n g (95% CI) I2 (95% PI NNT

CI)

Post-test

All comparisons 271 0.79 (0.70- 85 (83- –0.45 to 3.8

0.89) 86) 2.04

Outliers removed 194 0.70 (0.65- 26 (11- 0.49 to 4.4

0.74) 39) 0.90

Only low risk of bias 90 0.60 (0.49- 77 (72- –0.22 to 5.2

0.71) 81) 1.42

Three-level model 460 0.81 (0.72- 90 (-) –0.56 to 3.7

0.90) 2.17

Publication bias 349 0.47 (0.35- 90 (89- –1.52 to 7.0

correction 0.59) 91) 2.46

6-9 month follow-

up

All comparisons 78 0.74 (0.36- 91 (89- –1.90 to 4.1

1.11) 92) 3.37

Outliers removed 65 0.42 (0.33- 63 (51- –0.10 to 8.0

0.50) 72) 0.93

https://onlinelibrary.wiley.com/doi/full/10.1002/wps.21069 07/06/2025, 13 22
Page 14 of 32
:
 Only low risk of bias 29 0.91 (0.46- 94 (92- –1.46 to 3.2
PI – prediction interval, NNT – number needed to treat. The reported publication bias

correction is that using the trim and $ll procedure.

The sensitivity analyses supported the main $ndings (see Table 2 and
supplementary information). Heterogeneity was considerably lower after
excluding outliers (I2=26; 95% CI: 11-39), but the number of outliers that
had to be removed was large (n=77). The e"ect size was smaller for
studies with low risk of bias (g=0.60; 95% CI: 0.49-0.71) and after
adjusting for publication bias (g=0.47, 95% CI: 0.35-0.59 using the trim
and $ll procedure).

The subgroup analyses indicated that the e"ect size in studies with low
risk of bias was signi$cantly lower than in other studies (p<0.001), and
that the e"ect size di"ered across countries (higher in non-Western
countries; p=0.003) and treatment formats (higher for group formats;
p=0.02). There was no reduction of the e"ect size of CBT according to
the publication year (<2001 vs. 2001-2010 vs. >2011) (p=0.43). We
entered all variables in a multivariable meta-regression analysis and
found that, after adjustment for all variables, only the use of a waitlist
control condition (p=0.02) and whether the trial was conducted in an
“other” country (not the US, Europe, East Asia, Canada or Australia;
p=0.001) had a signi$cant impact on the e"ect size (see supplementary
information).

CBT was still e"ective at 6 to 9 month follow-up (g=0.74, 95% CI: 0.36-
1.11) and at 10 to 12 month follow-up (g=0.49, 95% CI: 0.01-0.98), and
this was con$rmed in most sensitivity analyses (see Table 2 and
supplementary information). Heterogeneity was high in most analyses.
At 13 to 24 month follow-up, the main e"ect size was no longer
signi$cant (g=0.22, 95% CI: –0.12 to 0.56), although this may be related to
the small number of studies (n=8).

https://onlinelibrary.wiley.com/doi/full/10.1002/wps.21069 07/06/2025, 13 22
Page 15 of 32
:
 The response rate was 0.42 (95% CI: 0.39-0.45) in CBT and 0.19 (95% CI:
0.18-0.21) in the control conditions, which resulted in a RR of 2.13 (95%
CI: 1.96-2.32) and a NNT of 4.7 (95% CI: 4.0-5.5) in favor of CBT (see
Table 3). Most sensitivity analyses indicated similar outcomes, except
that there was signi$cant publication bias, and the RR was lower in
studies with low risk of bias. The response rates di"ered signi$cantly
across control conditions, with the lowest rate for waitlist (see Table 3
and supplementary information).

Table 3. Cognitive behavior therapy (CBT) vs. control conditions:

response and remission rates, relative risk (RR) and number-needed-to-
treat (NNT)

n Rate I2 RR (95% I2 NNT

(95% CI) (95% CI) (95% (95% CI)
CI) CI)

Response

All CBT 238 0.42 82 2.13 47 4.7

conditions (0.39- (79- (1.96- (38- (4.0-
0.45) 84) 2.32) 54) 5.5)

Reported 10 0.42 91 2.32 46 4.0

(0.28- (85- (1.43- (0- (1.9-
0.59) 94) 3.77) 74) 12.2)

Estimated 228 0.42 81 2.13 47 4.7

(0.39- (79- (1.95- (38- (4.0-
0.45) 83) 2.32) 54) 5.5)

Outliers 162 0.42 31 2.25 10 4.2

excluded (0.40- (16- (2.07- (0- (3.7-
0.43) 43) 2.44) 25) 4.9)

https://onlinelibrary.wiley.com/doi/full/10.1002/wps.21069 07/06/2025, 13 22
Page 16 of 32
:
 Publication bias 259 0.39 84 1.66 59 8.0
correction (0.36- (82- (1.48- (54- (6.2-
0.42) 85) 1.85) 64) 11.0)

Low risk of bias 78 0.39 86 1.84 40 6.3

(0.35- (83- (1.64- (21- (4.9-

* di"erence among types of control conditions, p=0.006,

** di"erence between reported and estimated remission rates, p=0.02

The remission rate was 0.36 (95% CI: 0.31-0.42) for CBT and 0.15 (0.12-
0.18) for control conditions, which resulted in a RR of 2.45 (95% CI: 2.06-
2.92), and a NNT of 3.6 (95% CI: 2.7-5.0). This rate remained very similar
in the sensitivity analyses, although it was somewhat lower (but still
signi$cant) after adjustment for publication bias. These $ndings should
be considered with caution, because the di"erence between reported
and estimated remission rates was signi$cant (p=0.02) (see Table 3 and
supplementary information).

CBT versus other psychotherapies

CBT was compared with other psychotherapies in 87 studies (82
comparisons; 6,480 participants, including 3,148 in CBT and 3,332 in the
other therapies). The main analyses indicated a very small, but
signi$cant e"ect of CBT over other therapies (g=0.06; 95% CI: 0-0.12;
NNT=63), with low heterogeneity (I2=31; 95% CI: 10-47) (see Table 4).

Table 4. Cognitive behavior therapy (CBT) vs. other active treatments

n g (95% CI) I2 (95% NNT

CI)

CBT vs. other

https://onlinelibrary.wiley.com/doi/full/10.1002/wps.21069 07/06/2025, 13 22
Page 17 of 32
:
 psychotherapies

All studies 87 0.06 (0-0.12) 31 (10- 63

47)

Outliers removed 81 0.04 (–0.01 to 1 (0-27) 93.9

0.09)

Only low risk of bias 24 0.02 (–0.05 to 0 (0-45) 200.4

0.09)

Publication bias correction 92 0.04 (–0.03 to 44 (28- 93.4

0.11) 56)

Long-term e"ect (at 6-9 18 –0.03 (–0.14 to 0 (0-50) 117.2

months) 0.07)

Long-term e"ect (at 9-12 14 –0.09 (–0.19 to 12 (0- 47.7

months) 0.01) 50)

Compared to supportive 22 0.12 (–0.07 to 54 (26- 31.2

therapy 0.31) 72)

Compared to interpersonal 9 0.00 (–0.12 to 0 (0-65) 18.0

therapy 0.12)
NNT – number needed to treat. The reported publication bias correction is that using the

trim and $ll procedure.

When limiting the studies to those with low risk of bias, or excluding
outliers, or after adjustment for publication bias, the di"erence between
CBT and other psychotherapies was no longer signi$cant. In the
subgroup analyses in which we examined the di"erent psychotherapies
that were compared with CBT, we found no indication that one of these
therapies was more or less e"ective than CBT (see Table 4 and
supplementary information).

https://onlinelibrary.wiley.com/doi/full/10.1002/wps.21069 07/06/2025, 13 22
Page 18 of 32
:
 The number of studies reporting longer-term outcomes was small, and
no signi$cant di"erences between CBT and other psychotherapies were
found at 6-9 months, 9-12 months, or 13-24 months (see Table 4 and
supplementary information).

CBT versus pharmacotherapies and combined

treatment
CBT was compared with pharmacotherapies in 38 studies (38
comparisons; 2,979 participants, including 1,459 in CBT groups and
1,520 in pharmacotherapy groups). No signi$cant di"erence was found
between CBT and pharmacotherapies (g=0.08; 95% CI: –0.07 to 0.24).
The same was observed in sensitivity analyses, although one of the
analyses examining publication bias indicated a small, but signi$cant
e"ect in favor of pharmacotherapies. None of the subgroup analyses
pointed at a signi$cant di"erence between subgroups of studies (see
Table 4 and supplementary information).

At 6 to 12 month follow-up, CBT was more e"ective than

pharmacotherapies (g=0.34; 95% CI: 0.09-0.58; NNT=10.2). This was
con$rmed in most sensitivity analyses, although the number of studies
with low risk of bias was small and the e"ect size no longer signi$cant. In
two of the three analyses adjusting for publication bias, this $nding was
also not signi$cant anymore (see Table 4 and supplementary
information).

Combined treatment was compared with pharmacotherapy alone in 18

studies (18 comparisons; 1,658 participants, including 827 in the
combined and 831 in the pharmacotherapy conditions). Combined
treatment was more e"ective than pharmacotherapy (g=0.51; 95% CI:
0.19-0.84) and that was con$rmed in most sensitivity analyses, although
the number of trials with low risk of bias was small. After adjustment for
publication bias, the e"ects were no longer signi$cant. No signi$cant

https://onlinelibrary.wiley.com/doi/full/10.1002/wps.21069 07/06/2025, 13 22
Page 19 of 32
:
 di"erences were found in subgroup analyses (see Table 4 and
supplementary information).

Combined treatment was not signi$cantly more e"ective than CBT alone
(g=0.19; 95% CI: –0.11 to 0.50) in the 15 relevant studies (14
comparisons; 644 participants, including 325 in the combined and 319 in
the CBT only conditions). Only one of three analyses in which we
adjusted for publication bias resulted in a signi$cant e"ect size in favor
of combined treatment. Because of the limited number of trials, we
could only conduct a limited number of subgroup analyses, and none of
them resulted in signi$cant di"erences between subgroups (see Table 4
and supplementary information).

At 6 to 12 month follow-up, combined treatment was more e"ective

than pharmacotherapy alone (g=0.32, 95% CI: 0.09-0.55), but this $nding
was not con$rmed in all sensitivity analyses. Combined treatment was
not more e"ective than CBT alone (g=0.11; 95% CI: –0.38 to 0.60) (see
Table 4 and supplementary information).

Other comparisons
Unguided self-help CBT (Internet-based or not) had a small to moderate
e"ect on depression (g=0.45; 95% CI: 0.31-0.60), based on 36 studies (39
comparisons; 11,720 participants, including 6,206 in the CBT and 5,514 in
the control conditions). The e"ects of unguided CBT were signi$cant in
all sensitivity analyses, although they were somewhat smaller in two of
three analyses adjusting for publication bias. Subgroup analyses
indicated that waitlist-controlled trials resulted in larger e"ect sizes
(p=0.03), and studies in Europe resulted in smaller e"ects (p=0.01). We
also found that studies conducted after 2011 had signi$cantly larger
e"ects than earlier studies (p=0.01), suggesting that the e"ects may have
improved over time (see Table 5 and supplementary information).

https://onlinelibrary.wiley.com/doi/full/10.1002/wps.21069 07/06/2025, 13 22
Page 20 of 32
:
 Table 5. Other comparisons between cognitive behavior therapy (CBT)
and control conditions

n g (95% CI) I2 (95% CI) NNT

Unguided self-help CBT

All comparisons 39 0.45 (0.31-0.60) 78 (71- 7.2

84)

Outliers removed 34 0.43 (0.34-0.52) 51 (28- 7.7

67)

Only low risk of bias 18 0.40 (0.27-0.52) 59 (32- 8.4

76)

Publication bias correction 53 0.25 (0.07-0.43) 84 (80- 14.2

88)

CBT in institutional settings

All comparisons 11 0.65 (0.21-1.08) 70 (45- 4.8

84)

Outliers removed 10 0.49 (0.15-0.83) 52 (2-77) 6.6

Publication bias correction 13 0.41 (–0.14 to 81 (68- 8.2

0.96) 88)

CBT in children and

adolescents

All comparisons 39 0.41 (0.25-0.57) 78 (70- 8.1

NNT – number needed to treat. The reported publication bias correction is that using the

trim and $ll procedure.

https://onlinelibrary.wiley.com/doi/full/10.1002/wps.21069 07/06/2025, 13 22
Page 21 of 32
:
 We could compare CBT in institutional settings to control conditions in
10 studies (11 comparisons; 448 participants, including 275 in CBT and
173 in the control conditions). Five studies (six comparisons) were
conducted in psychiatric inpatient settings, four in nursing homes, and
one in another institutional setting. None of the trials was rated as at low
risk of bias. We found a moderate to large e"ect (g=0.65; 95% CI: 0.21-
1.08) with high heterogeneity, which remained signi$cant in most
sensitivity analyses, but was no longer signi$cant in two of the three
analyses adjusting for publication bias (see Table 5 and supplementary
information). Because of the small number of trials and the low quality,
we did not conduct subgroup analyses.

In children and adolescents, CBT was compared to control conditions in

37 studies (39 comparisons; 3,667 participants, including 1,859 in CBT
and 1,808 in control groups). We found a moderate e"ect (g=0.41; 95%
CI: 0.25-0.57; NNT=8.1), with high heterogeneity (I2=78; 95% CI: 70-84).
The e"ect size remained similar across most sensitivity analyses. The
number of studies with low risk of bias was low and the e"ect size was
no longer signi$cant in this subset. One of the e"ect sizes adjusted for
publication bias was also not signi$cant (see Table 5 and supplementary
information). In the subgroup analyses, we found that waitlist control
groups resulted in signi$cantly larger e"ect sizes than other control
conditions (p=0.01), and studies with low risk of bias resulted in
signi$cantly lower e"ect sizes than other studies (p=0.04).

DISCUSSION
This is the largest meta-analysis ever of a speci$c type of psychotherapy
for a mental disorder, including 409 RCTs (518 comparisons) with 52,702
patients. CBT was found to be e"ective in depression when compared to
control conditions such as usual care and waitlist, with a moderate to
large e"ect size (g=0.79). This e"ect was robust in several sensitivity

https://onlinelibrary.wiley.com/doi/full/10.1002/wps.21069 07/06/2025, 13 22
Page 22 of 32
:
 analyses, although it was somewhat smaller for studies with low risk of
bias (g=0.60) and after adjustment for publication bias (g=0.47). CBT was
still signi$cantly e"ective at 6-9 month (g=0.74) and 10-12 month
(g=0.49) follow-up, and this was con$rmed in most sensitivity analyses.

A total of 42% of patients receiving CBT responded to treatment, while

the response rate was only 19% in control groups, with a NNT of 4.7 in
favor of CBT. The remission rate was 36% in patients receiving CBT,
compared to 15% in control conditions, with a NNT of 3.6.

Comparative trials suggest that CBT is signi$cantly more e"ective than

other psychotherapies, but the di"erence is small (g=0.06) and does not
remain signi$cant in most sensitivity analyses. The e"ects of CBT are
comparable to those of pharmacotherapies at the short term, but CBT is
signi$cantly more e"ective at 6 to 12 months (g=0.34). Combined
treatment is signi$cantly more e"ective than pharmacotherapy alone, at
the short (g=0.51) and the longer term (g=0.32), but combined treatment
is not more e"ective than CBT alone at either time point.

Most trials examine CBT in an individual, group or guided self-help

format, and we previously showed that there are no signi$cant
di"erences between these formats12. In the current meta-analysis, we
could also include a set of trials of unguided self-help CBT, and found
that this was also e"ective, with a small to moderate e"ect size (g=0.45).
CBT was also found to be e"ective in inpatient settings (g=0.65), as well
as in children and adolescents (g=0.41).

Research on CBT has evolved over time. The quality of studies has
improved, which can be seen from the increasing number of trials with
low risk of bias, the decrease in the use of waitlist control groups, and
the increase in sample sizes of included studies. The number of
treatment sessions has signi$cantly decreased over the years. In a meta-
regression analysis, we could not con$rm that the e"ect size of CBT has
37

https://onlinelibrary.wiley.com/doi/full/10.1002/wps.21069 07/06/2025, 13 22
Page 23 of 32
:
 decreased over time, as was suggested in an earlier study37.

The $ndings of this study should be considered in the light of some

limitations. First, heterogeneity was high in many analyses, and
subgroup and meta-regression analyses could not identify all sources of
this heterogeneity, suggesting that there are di"erences between trials
that cannot be explained by the extracted characteristics. Second, risk of
bias was high in many of the included trials, and the e"ect sizes of the
trials with low risk of bias were signi$cantly lower in some of the
analyses. Fortunately, the number of studies was so large that we could
examine outcomes in subsets of trials with low risk of bias. Finally, we
found indications of publication bias in many analyses, although several
$ndings remained robust after correcting for this bias.

We can conclude that CBT is e"ective in the treatment of depression

with a moderate to large e"ect size, and that its e"ect is still signi$cant
up to 12 months. The superiority of CBT over other psychotherapies
does not emerge clearly from this meta-analysis. CBT appears to be as
e"ective as pharmacotherapies at the short term, but more e"ective at
the longer term. Combined treatment appears to be superior to
pharmacotherapy alone but not to CBT alone. The e#cacy of CBT in
depression is documented across di"erent delivery formats, ages, target
groups, and settings.

ACKNOWLEDGEMENTS
Supplementary information on this study is available at
http://osf.io/a6p3w.

REFERENCES 

1 World Health Organization. World mental health report: transforming

https://onlinelibrary.wiley.com/doi/full/10.1002/wps.21069 07/06/2025, 13 22
Page 24 of 32
:
 mental health for all. Geneva: World Health Organization, 2022.

Google Scholar

2 Herrman H, Patel V, Kieling C et al. Time for united action on depression:

a Lancet-World Psychiatric Association Commission. Lancet 2022; 399:
957-1022.

PubMed Web of Science® Google Scholar

3 Cuijpers P, Vogelzangs N, Twisk J et al. Comprehensive meta-analysis of

excess mortality in depression in the general community versus patients
with speci$c illnesses. Am J Psychiatry 2014; 171: 453-62.

PubMed Web of Science® Google Scholar

4 Chisholm D, Sweeny K, Sheehan P et al. Scaling-up treatment of

depression and anxiety: a global return on investment analysis. Lancet
Psychiatry 2016; 3: 415-24.

PubMed Web of Science® Google Scholar

5 Cipriani A, Furukawa TA, Salanti G et al. Comparative e#cacy and

acceptability of 21 antidepressant drugs for the acute treatment of adults
with major depressive disorder: a systematic review and network meta-
analysis. Lancet 2018; 391: 1357-66.

CAS PubMed Web of Science® Google Scholar

6 Cuijpers P, Quero S, Noma H et al. Psychotherapies for depression: a

network meta-analysis covering e#cacy, acceptability and long-term
outcomes of all main treatment types. World Psychiatry 2021; 20: 283-93.

PubMed Web of Science® Google Scholar

https://onlinelibrary.wiley.com/doi/full/10.1002/wps.21069 07/06/2025, 13 22
Page 25 of 32
:
 7 Cuijpers P, Karyotaki E, de Wit L et al. The e"ects of $fteen evidence-
supported therapies for adult depression: a meta-analytic review.
Psychother Res 2020; 30: 279-93.

PubMed Web of Science® Google Scholar

8 Cuijpers P, Berking M, Andersson G et al. A meta-analysis of cognitive

behavior therapy for adult depression, alone and in comparison to other
treatments. Can J Psychiatry 2013; 58: 376-85.

PubMed Web of Science® Google Scholar

9 Cuijpers P, Noma H, Karyotaki E et al. A network meta-analysis of the

e"ects of psychotherapies, pharmacotherapies and their combination in
the treatment of adult depression. World Psychiatry 2020; 19: 92-107.

PubMed Web of Science® Google Scholar

10 Cuijpers P, Karyotaki E. A meta-analytic database of randomised trials

on psychotherapies for depression. https://doi.org/10.17605/OSF.IO/825C
6.

Google Scholar

11 Cuijpers P, van Straten A, Andersson G et al. Psychotherapy for

depression in adults: a meta-analysis of comparative outcome studies. J
Consult Clin Psychol 2008; 76: 909-22.

PubMed Web of Science® Google Scholar

12 Cuijpers P, Noma H, Karyotaki E et al. E"ectiveness and acceptability of

cognitive behavior therapy delivery formats in adults with depression: a
network meta-analysis. JAMA Psychiatry 2019; 76: 700-7.

https://onlinelibrary.wiley.com/doi/full/10.1002/wps.21069 07/06/2025, 13 22
Page 26 of 32
:
 PubMed Web of Science® Google Scholar

13 Cuijpers P, Ciharova M, Miguel C et al. Psychological treatment of

depression in institutional settings: a meta-analytic review. J A!ect Disord
2021; 286: 340-50.

PubMed Web of Science® Google Scholar

14 Higgins JPT, Altman DG, Gøtzsche PC et al. The Cochrane

Collaboration's tool for assessing risk of bias in randomised trials. BMJ
2011; 343:d5928.

PubMed Web of Science® Google Scholar

15 Sterne JAC, Savovic J, Page MJ et al. RoB 2: a revised tool for assessing
risk of bias in randomised trials. BMJ 2019; 366:l4898.

PubMed Google Scholar

16 Hedges LV, Olkin I. Statistical methods for meta-analysis. San Diego:

Academic Press, 1985.

CAS Google Scholar

17 Harrer M, Kuper P, Cuijpers P. metapsyTools: several R helper

functions for the “metapsy” database. R package version 0.3.2, 2022. http
s://tools.metapsy.org.

Google Scholar

18 Balduzzi S, Rücker G, Schwarzer G. How to perform a meta-analysis

with r: a practical tutorial. Evid Based Ment Health 2019; 22: 153-60.

https://onlinelibrary.wiley.com/doi/full/10.1002/wps.21069 07/06/2025, 13 22
Page 27 of 32
:
 PubMed Web of Science® Google Scholar

19 Viechtbauer W. Conducting meta-analyses in r with the metafor

package. J Statist Softw 2010; 36: 1-48.

Web of Science® Google Scholar

20 Harrer M, Cuijpers P, Furukawa T et al. dmetar: companion R package

for the guide ‘doing meta-analysis in R’. R package version 0.0.9000. http://
dmetar.protectlab.org.

Google Scholar

21 Pustejovsky JE, Tipton E. Meta-analysis with robust variance estimation:

expanding the range of working models. Prev Sci 2022; 23: 425-38.

PubMed Web of Science® Google Scholar

22 Viechtbauer W, Cheung MWL. Outlier and in%uence diagnostics for

meta-analysis. Res Synth Meth 2010; 1: 112-25.

PubMed Web of Science® Google Scholar

23 Maier M, VanderWeele TJ, Mathur MB. Using selection models to

assess sensitivity to publication bias: a tutorial and call for more routine
use. Campbell Syst Rev 2022; 18:e1256.

PubMed Web of Science® Google Scholar

24 Duval S, Tweedie R. Trim and $ll: a simple funnel-plot-based method of

testing and adjusting for publication bias in meta-analysis. Biometrics 2000;
56: 455-63.

https://onlinelibrary.wiley.com/doi/full/10.1002/wps.21069 07/06/2025, 13 22
Page 28 of 32
:
 CAS PubMed Web of Science® Google Scholar

25 Rücker G, Schwarzer G, Carpenter JR et al. Treatment-e"ect estimates

adjusted for small-study e"ects via a limit meta-analysis. Biostatistics 2011;
12: 122-42.

PubMed Web of Science® Google Scholar

26 McShane BB, Böckenholt U, Hansen KT. Adjusting for publication bias

in meta-analysis: an evaluation of selection methods and some cautionary
notes. Perspect Psychol Sci 2016; 11: 730-49.

PubMed Web of Science® Google Scholar

27 Carter EC, Schönbrodt FD, Gervais WM et al. Correcting for bias in

psychology: a comparison of meta-analytic methods. Adv Meth Pract Psychol
Sci 2019; 2: 115-44.

Google Scholar

28 IntHout J, Ioannidis JP, Borm GF. The Hartung-Knapp-Sidik-Jonkman

method for random e"ects meta-analysis is straightforward and
considerably outperforms the standard DerSimonian-Laird method. BMC
Med Res Methodol 2014; 14: 25.

PubMed Web of Science® Google Scholar

29 Higgins JPT, Thompson SG, Deeks J et al. Measuring inconsistency in

meta-analyses. BMJ 2003; 327: 557-60.

PubMed Google Scholar

30 Cheung MWL. Modeling dependent e"ect sizes with three-level meta-

https://onlinelibrary.wiley.com/doi/full/10.1002/wps.21069 07/06/2025, 13 22
Page 29 of 32
:
 analyses: a structural equation modeling approach. Psychol Methods 2014;
19: 211-29.

PubMed Web of Science® Google Scholar

31 Borenstein M, Hedges LV, Higgins JPT et al. Introduction to meta-

analysis. Chichester: Wiley, 2009.

CAS PubMed Google Scholar

32 Borenstein M, Higgins JPT, Hedges LV et al. Basics of meta-analysis: I2 is

not an absolute measure of heterogeneity. Res Syn Meth 2017; 8: 5-18.

PubMed Web of Science® Google Scholar

33 Furukawa TA. From e"ect size into number needed to treat. Lancet
1999; 353: 1680.

CAS PubMed Web of Science® Google Scholar

34 Cuijpers P, Karyotaki E, Ciharova M et al. The e"ects of

psychotherapies for depression on response, remission, reliable change,
and deterioration: a meta-analysis. Acta Psychiatr Scand 2021; 144: 288-99.

PubMed Web of Science® Google Scholar

35 Furukawa TA, Cipriani A, Barbui C et al. Imputing response rates from

means and standard deviations in metaanalyses. Int Clin Psychopharmacol
2005; 20: 49-52.

PubMed Web of Science® Google Scholar

36 Frank E, Prien RF, Jarrett RB et al. Conceptualization and rationale for

https://onlinelibrary.wiley.com/doi/full/10.1002/wps.21069 07/06/2025, 13 22
Page 30 of 32
:
 consensus de$nitions of terms in major depressive disorder: remission,
recovery, relapse, and recurrence. Arch Gen Psychiatry 1991; 48: 851-5.

CAS PubMed Web of Science® Google Scholar

37 Johnsen TJ, Friborg O. The e"ects of cognitive behavioral therapy as an

anti-depressive treatment is falling: a meta-analysis. Psychol Bull 2015; 141:
747-68.

PubMed Web of Science® Google Scholar

Citing Literature 

Download PDF

ABOUT WILEY ONLINE LIBRARY

Privacy Policy
Terms of Use
About Cookies
Manage Cookies
Accessibility
Wiley Research DE&I Statement and Publishing Policies
Developing World Access

HELP & SUPPORT

Contact Us
Training and Support

https://onlinelibrary.wiley.com/doi/full/10.1002/wps.21069 07/06/2025, 13 22
Page 31 of 32
:
 DMCA & Reporting Piracy

OPPORTUNITIES

Subscription Agents
Advertisers & Corporate Partners

CONNECT WITH WILEY

The Wiley Network

Wiley Press Room

Copyright © 1999-2025 John Wiley & Sons, Inc or related

companies. All rights reserved, including rights for text and
data mining and training of arti!cial intelligence technologies or
similar technologies.

https://onlinelibrary.wiley.com/doi/full/10.1002/wps.21069 07/06/2025, 13 22
Page 32 of 32
: